Re: [patch, mips] More mips memcpy improvements

[Steve Ellcey <sellcey at mips dot com>]

On Wed, 2012-11-28 at 09:46 -0500, Carlos O'Donell wrote:
> On Wed, Nov 28, 2012 at 2:43 AM, Maxim Kuvyrkov <maxim@codesourcery.com> wrote:
> > On 28/11/2012, at 9:37 AM, Steve Ellcey wrote:
> >
> > ...
> >> Andrew and Maxim, could you take a look at this new version and see if it
> >> works OK for you?
> >
> > Correctness-wise, the patch does not trigger any faults on the benchmark that I'm using, so that's good.  Performance-wise, I didn't see any significant difference.  It may be that XLP's cache doesn't benefit from using prepare-for-store hints.  In any case, the patch doesn't regress performance, so I'm OK with it.
> >
> > Steve, did you run normal glibc testsuite on this patch with no failures?
> 
> I will note that I'm not comfortable pushing a patch like this so
> close to the 2.17 freeze.
> 
> My preference would be that 2.17 goes out the door, then once the
> freeze is over we put this patch into trunk and then everyone can test
> it and ensure that what we get is continual incremental improvement.
> 
> Cheers,
> Carlos.

I can wait to do the checkin if you want.  I would point out that the
current MIPS memcpy was only checked in a month ago so it doesn't have
much more testing then this version.  I haven't run the glibc testsuite
with this memcpy, I used a combination of memcpy specific tests I wrote
plus GCC testing for verification.  I will run the glibc testsuite
today, it hasn't been clean in the past for me (before I made any
changes) and I don't expect it to be so now but I will report on my
results when I have them.

Steve Ellcey
sellcey@mips.com